On 20/05/2020 18:00, François Ozog wrote:
On Wed, 20 May 2020 at 18:39, Daniel Thompson <daniel.thompson@linaro.org mailto:daniel.thompson@linaro.org> wrote:
On Tue, May 19, 2020 at 03:29:01PM +0100, Grant Likely wrote: > On 11/03/2020 16:42, Daniel Thompson wrote: > > On Wed, Mar 11, 2020 at 01:42:36PM +0100, Francois Ozog wrote: > > > We have the following cases: > > > > > > - FW compatible with GPT (I mean firmware can be searched based on > > > GUID partition) > > > Ok > > > > > > - FW that uses offsets and can be positioned at LBA >= 33 > > > Ok > > > Need to define a protective partition >> > > > - FW that uses offsets and can be positioned such that space between LBA-2 > > > and LBA-33 is used. > > > Ok in theory as the header states where the partition entries location is > > > specified in a GPT_HEADER "Starting LBA of array of partition entries". > > > Linux kernel properly loads the partition entries if we push them after > > > 16MB. > > > > > > read_lba <https://elixir.bootlin.com/linux/latest/ident/read_lba>(state, > > > le64_to_cpu <https://elixir.bootlin.com/linux/latest/ident/le64_to_cpu>(gpt->partition_entry_lba) > > > > > > But I bet 2 is hardcoded in many tools... > > > > Agree... but that's "just bugs" and I suspect we could get >90% test > > coverage for Linux systems just by checking util-linux (and the kernel > > itself). Maybe for extra style points also check on of the BSDs. > > It is worth stepping back from the details to take a look at the intent. The > purpose of this entire section of EBBR is to describe how firmware and the > OS can co-exist on the same media device. In broad strokes it means if > firmware is stored on the block device, then the OS must constrain how it > uses the device. > > On platforms with separate firmware storage (e.g., SPI flash or UFS boot > partitions) this isn't an issue. The OS can blow away everything on the disk > and recreate it. > > But when it is an issue, the rules need to lay down what regions (offsets, > partitions, or file paths) firmware is allowed to own and what the OS is and > is not allowed to do. e.g., the OS is allowed to erase and recreate the OS > partitions, but it is not allowed to write a blank GPT or erase the system > partition. > > I think the EBBR spec should focus on defining exactly what restrictions on > the OS are, and how the restrictions are communicated. Then OS vendors have > a fighting chance of supporting the restricted platforms well. > > Ultimately though this is a guide and the OS could choose to ignore the > restrictions... in which case it gets to keep the unbootable brick when it > does. :-) Agree with all above. Also I think we can turn at least part of the original issue into a concrete question. We have a SOC with some magic values hard coded into its boot ROM. The System Firmware author wants to ship it with the following GPT on the shared eMMC. LBA0 Protective MBR LBA1 Primary GPT header LBA2..18001 Reserved, mixture of dead space and a system firmware loaded by Boot rom LBA18002 Start of partition arrray (Entry 1, 2, 3, 4) ... LBA18033 End of partition arrray LBA18034 Start of allocatable partition space LBA-33..-1 End of disk is labelled as normal (or in a shorter GPT jargon form, a system where PartitionEntryLBA is 18002). Is such a system EBBR compliant? If yes, should it be?I would say it is not EBBR compliant because it does not follow EFI spec and we mandate it in EBBR.
Not quite that cut and dry. We have the ability to specify exceptions to what UEFI specifies where it is warranted to do so.
In this particular example the GPT looks like it is entirely spec-compliant, but if the OS recreates the GPT without accounting for the firmware space then it results in a dead platform. Within spec, but with behaviour an OS might not be prepared for. In this case EBBR should specify how the OS knows not to mess with the GPT layout.
g.
Is the use case "valid"? I think it is valid because when you deal with immutable BootROM you don't want complex code, GPT may evolve so that you would have to evolve the BootROM... If we conclude this is a valid use case (and not creating ugly legacy to deal with in the future), we need a clean reservation in EFI so that GPT can start at an arbitrary LBA as 18002. enhancing the protective MBR semantics does not seem too complex to achieve. Can we list SoCs that have similar characteristic?
Daniel. PS 18002 is arbitrary but I think the example is sufficient in this form and it was easier to diagram with a concrete number.-- François-Frédéric Ozog | /Director Linaro Edge & Fog Computing Group/ T: +33.67221.6485 francois.ozog@linaro.org mailto:francois.ozog@linaro.org | Skype: ffozog